In this chapter, students will be able to understand: Introduction, the use of intercept dummy variables, slope dummy variables, an example: the university effect on house prices, common applications of dummy variables, testing for the existence of qualitative effects, testing the equivalence of two regressions using dummy variables.
Chapter Dummy (Binary) Variables 9.1 Introduction The multiple regression model yt = β1 + β2xt2 + β3xt3 +…+ βKxtK + et (9.1.1) The assumptions of the multiple regression model are Assumptions of the Multiple Regression Model MR1 yt = β1 + β2xt2 + β3xt3 +…+ βKxtK + et, t = 1,…,T MR2 E(yt) = β1 + β2xt2 + β3xt3 +…+ βKxtK ⇔ E(et) = MR3 var(yt) = var(et) = σ2 Slide 9.1 Undergraduate Econometrics, 2nd Edition –Chapter MR4 cov(yt, ys) = cov(et, es) = MR5 The values of xtK are not random and are not exact linear functions of the other explanatory variables MR6 yt ~ N[(β1 + β2xt2 + β3xt3 +…+ βKxtK), σ2] ⇔ et ~ N(0, σ2) • Assumption MR1 defines the statistical model that we assume is appropriate for all T of the observations in our sample One part of the assertion is that the parameters of the model, βK, are the same for each and every observation Recall that βK = the change in E(yt) when xtK is increased by one unit, and all other variables are held constant = ∆E ( yt ) ∂E ( yt ) = ∆xtk (other variables held constant) ∂xtk Slide 9.2 Undergraduate Econometrics, 2nd Edition –Chapter • Assumption implies that for each of the observations t = 1, ,T the effect of a one unit change in xtK on E(yt) is exactly the same If this assumption does not hold, and if the parameters are not the same for all the observations, then the meaning of the least squares estimates of the parameters in Equation (9.1.1) is not clear • In this chapter we extend the multiple regression model of Chapter to situations in which the regression parameters are different for some of the observations in a sample We use dummy variables, which are explanatory variables that take one of two values, usually or These simple variables are a very powerful tool for capturing qualitative characteristics of individuals, such as gender, race, and geographic region of residence In general, we use dummy variables to describe any event that has only two possible outcomes We explain how to use dummy variable to account for such features in our model • As a second tool for capturing parameter variation, we make use of interaction variables These are variables formed by multiplying two or more explanatory Slide 9.3 Undergraduate Econometrics, 2nd Edition –Chapter variables together When using either dummy variables or interaction variables, some changes in model interpretation are required We will discuss each of these scenarios Slide 9.4 Undergraduate Econometrics, 2nd Edition –Chapter 9.2 The Use of Intercept Dummy Variables Dummy variables allow us to construct models in which some or all regression model parameters, including the intercept, change for some observations in the sample To make matters specific, we consider an example from real estate economics Buyers and sellers of homes, tax assessors, real estate appraisers, and mortgage bankers are interested in predicting the current market value of a house A common way to predict the value of a house is to use a “hedonic” model, in which the price of the house is explained as a function of its characteristics, such as its size, location, number of bedrooms, age, etc • For the present, let us assume that the size of the house, S, is the only relevant variable in determining house price, P Specify the regression model as Pt = β1 + β2St + et (9.2.1) Slide 9.5 Undergraduate Econometrics, 2nd Edition –Chapter In this model β2 is the value of an additional square foot of living area, and β1 is the value of the land alone • Dummy variables are used to account for qualitative factors in econometric models They are often called binary or dichotomous variables as they take just two values, usually or 0, to indicate the presence or absence of a characteristic That is, a dummy variable D is 1 D= 0 if characteristic is present if characteristic is not present (9.2.2) Thus, for the house price model, we can define a dummy variable to account for a desirable neighborhood, as Slide 9.6 Undergraduate Econometrics, 2nd Edition –Chapter 1 Dt = 0 if property is in the desirable neighborhood if property is not in the desirable neighborhood (9.2.3) • Adding this variable to the regression model, along with a new parameter δ, we obtain Pt = β1 + δDt + β2St + et (9.2.4) • The effect of the inclusion of a dummy variable Dt into the regression model is best seen by examining the regression function, E(Pt), in the two locations If the model in (9.2.4) is correctly specified, then E(et) = and (β + δ) + β2 St E ( Pt ) = β1 + β2 St when Dt = when Dt = (9.2.5) Slide 9.7 Undergraduate Econometrics, 2nd Edition –Chapter • In the desirable neighborhood, Dt = 1, and the intercept of the regression function is (β1 + δ) In other areas the regression function intercept is simply β1 This difference is depicted in Figure 9.1, assuming that δ > • Adding the dummy variable Dt to the regression model creates a parallel shift in the relationship by the amount δ In the context of the house price model the interpretation of the parameters δ is that it is a “location premium,” the difference in house price due to being located in the desirable neighborhood • A dummy variable like Dt that is incorporated into a regression model to capture a shift in the intercept as the result of some qualitative factor is an intercept dummy variable In the house price example we expect the price to be higher in a desirable location, and thus we anticipate that δ will be positive • The least squares estimator’s properties are not affected by the fact that one of the explanatory variables consists only of zeros and ones Dt is treated as any other explanatory variable We can construct an interval estimate for δ, or we can test the Slide 9.8 Undergraduate Econometrics, 2nd Edition –Chapter significance of its least squares estimate Such a test is a statistical test of whether the neighborhood effect on house price is “statistically significant.” If δ = 0, then there is no location premium for the neighborhood in question Slide 9.9 Undergraduate Econometrics, 2nd Edition –Chapter 9.3 Slope Dummy Variables • We can allow for a change in a slope by including in the model an additional explanatory variable that is equal to the product of a dummy variable and a continuous variable In our model the slope of the relationship is the value of an additional square foot of living area If we assume this is one value for homes in the desirable neighborhood, and another value for homes in other neighborhoods, we can specify Pt = β1 + β2St + γ(StDt) + et (9.3.1) • The new variable (StDt) is the product of house size and the dummy variable, and is called an interaction variable, as it captures the interaction effect of location and size on house price Alternatively, it is called a slope dummy variable, because it allows for a change in the slope of the relationship Slide 9.10 Undergraduate Econometrics, 2nd Edition –Chapter The efficacy of the investment tax credit program is checked by testing the null hypothesis that δ = against the alternative that δ ≠ 0, or δ > 0, using the appropriate two- or one-tailed t-test 9.6.2 Testing Jointly for the Presence of Several Qualitative Effects • If a model has more than one dummy variable, representing several qualitative characteristics, the significance of each, apart from the others, can be tested using the t-test outlined in the previous section If it is often of interest, however, to test the joint significance of all the qualitative factors • For example, consider the wage Equation (9.5.1) WAGE = β1 + β2EXP + δ1RACE + δ2SEX + γ(RACE × SEX) + e (9.6.1) Slide 9.32 Undergraduate Econometrics, 2nd Edition –Chapter How we test the hypothesis that neither race nor gender affects wages? We it by testing the joint null hypothesis H0: δ1 = 0, δ2 = 0, γ = against the alternative that at least one of the indicated parameters is not zero If the null hypothesis is true, race and gender fall out of the regression, and thus have no effect on wages • To test this hypothesis we use the F-test procedure that is described in Chapter 8.1 The test statistic for a joint hypothesis is F= ( SSER − SSEU ) / J SSEU /(T − K ) (9.6.2) where SSER is the sum of squared least squares residuals from the “restricted” model in which the null hypothesis is assumed to be true, SSEU is the sum of squared residuals from the original, “unrestricted,” model, J is the number of joint hypotheses, and (T − K) is the number of degrees of freedom in the unrestricted model Slide 9.33 Undergraduate Econometrics, 2nd Edition –Chapter • If the null hypothesis is true, then the test statistic F has an F-distribution with J numerator degrees of freedom and (T − K) denominator degrees of freedom, F(J, T-K) We reject the null hypothesis if F ≥ Fc, where Fc is the critical value for the level of significance α • In order to test the J = joint null hypotheses H0: δ1 = 0, δ2 = 0, γ = 0, we obtain the unrestricted sum of squared errors SSEU by estimating Equation (9.6.1) The restricted sum of squares SSER is obtained by estimating the restricted model WAGE = β1 + β2EXP + e (9.6.3) Slide 9.34 Undergraduate Econometrics, 2nd Edition –Chapter 9.7 Testing the Equivalence of Two Regressions Using Dummy Variables • In Equation (9.3.3) we assume that house location affects both the intercept and the slope The resulting regression model is Pt = β1 + δDt + β2St + γ(StDt) + et (9.7.1) The regression functions for the house prices in the two locations are (β + δ) + (β2 + γ ) St = α1 + α St E ( Pt ) = β1 + β2 St desirable neighborhood data other neighborhood data (9.7.2) • Note that since we have allowed the intercept and slope to differ, we have essentially assumed that the regressions in the two neighborhoods are completely different We Slide 9.35 Undergraduate Econometrics, 2nd Edition –Chapter can apply least squares separately to data from the two neighborhoods to obtain estimates of α1 and α2, and β1 and β2, in Equation (9.7.2) 9.7.1 The Chow Test • An important question is “Are there differences between the hedonic regressions for the two neighborhoods or not?” If there are no differences, then the data from the two neighborhoods can be “pooled” into one sample, with no allowance made for differing slope or intercept • If the joint null hypothesis H0: δ = 0, γ = is true, then there are no differences between the base price and price per square foot in the two neighborhoods If we reject this null hypothesis then the intercepts and/or slopes are different, we cannot simply pool the data and ignore neighborhood effects • From Equation (9.7.2), by testing H0: δ = 0, γ = we are testing the equivalence of the two regressions Slide 9.36 Undergraduate Econometrics, 2nd Edition –Chapter Pt = α1 + α2St + et Pt = β1 + β2St + et (9.7.3) • If δ = then α1 = β1, and if γ = 0, then α2 = β2 In this case we can simply estimate the “pooled” Equation (9.2.1), Pt = β1 + β2St + et, using data from the two neighborhoods together • If we reject either or both of these hypotheses, then the equalities α1 = β1 and α2 = β2 are not true, in which case pooling the data together would be equivalent to imposing constraints, or restrictions, which are not true • Testing the equivalence of two regressions is sometimes called a Chow test, after econometrician Gregory Chow, who studied some aspects of this type of testing We carry out the test by creating an intercept and slope dummy for every variable in the Slide 9.37 Undergraduate Econometrics, 2nd Edition –Chapter model, and then jointly testing the significance of the dummy variable coefficients using an F-test Slide 9.38 Undergraduate Econometrics, 2nd Edition –Chapter 9.7.2 An Empirical Example of The Chow Test • As an example, let us consider the investment behavior of two large corporations, General Electric and Westinghouse These firms compete against each other and produce many of the same types of products We might wonder if they have similar investment strategies In Table 9.3 are investment data for the years 1935 to 1954 (this is a classic data set) for these two corporations The variables, for each firm, in 1947 dollars, are INV = gross investment in plant and equipment V = value of the firm = value of common and preferred stock K = stock of capital Slide 9.39 Undergraduate Econometrics, 2nd Edition –Chapter A simple investment function is INVt = β1 + β2Vt + β3Kt + et (9.7.4) • If we combine, or pool, the data for both firms we have T = 40 observations with which to estimate the parameters of the investment function But pooling the two sources of data is valid only if the regression parameters and the variances of the error terms are the same for both corporations If these parameters are not the same, and we combine the data sets anyway, it is equivalent to restricting the investment functions of the two firms to be identical when they are not, and the least squares estimators of the parameters in the restricted model (9.7.4) are biased and inconsistent Estimating the restricted, pooled, model by least squares provides the restricted sum of squared errors, SSER, that we will use in the formation of an F-test statistic Slide 9.40 Undergraduate Econometrics, 2nd Edition –Chapter • Using the Chow test we can test whether or not the investment functions for the two firms are identical To so, let D be a dummy variable that is for the 20 Westinghouse observations, and otherwise We then include an intercept dummy variable and a complete set of slope dummy variables INVt = β1 + δ1Dt + β2Vt + δ2(DtVt) + β3Kt + δ3(DtKt) + et (9.7.5) This is an unrestricted model From the least squares estimation of this model we will obtain the unrestricted sum of squared errors, SSEU, that we will use in the construction of an F-statistic shown in Equation (9.6.2) • We test the equivalence of the investment regression functions for the two firms by testing the J = joint null hypotheses H0: δ1 = 0, δ2 = 0, δ3 = against the alternative H1: at least one δi ≠ Slide 9.41 Undergraduate Econometrics, 2nd Edition –Chapter • Using the data in Table 9.3, the estimated restricted and unrestricted models, with tstatistics in parentheses, and their sums of squared residuals are as follows Restricted (one relation for all observations) Model: ˆ = 17.8720 + 0.0152V + 0.1436 K INV (2.544) (2.452) (7.719) (9.7.6) SSER =16563.00 Slide 9.42 Undergraduate Econometrics, 2nd Edition –Chapter Unrestricted Model: ˆ = −9.9563 + 9.4469 D + 0.0266V INV (0.421) (0.328) (2.265) + 0.0263( D × V ) + 0.1517 K − 0.0593( D × K ) (0.767) (7.837) ( − 0.507) (9.7.7) SSEU = 14989.82 Constructing the F-statistic, F= ( SSER − SSEU ) / J (16563.00 − 14989.82) / = = 1.1894 SSEU /(T − K ) 14989.82 /(40 − 6) (9.7.8) Slide 9.43 Undergraduate Econometrics, 2nd Edition –Chapter • The α = 05 critical value Fc = 2.8826 comes from the F(3, 34) distribution Since F < Fc we can not reject the null hypothesis that the investment functions for General Electric and Westinghouse are identical • In this case the joint F-test and the individual t-tests of the dummy variable and slope dummy variables reach the same conclusion However, remember that the t- and Ftests have different purposes and their outcomes will not always match in this way • It is interesting that for the Chow test we can calculate SSEU, the unrestricted sum of squared errors another way, which is frequently used in practice Instead of estimating the model (9.7.5) to obtain SSEU, we can estimate the simpler model in (9.7.4) twice Using the T = 20 General Electric observations estimate (9.7.4) by least squares; call the sum of squared residuals from this estimation SSE1 Then, using the T = 20 Westinghouse observations, estimate (9.7.4) by least squares; call the sum of squared residuals from this estimation SSE2 The unrestricted sum of squared residuals SSEU from (9.7.5) is identical to the sum SSE1 + SSE2 The advantage of this approach to Slide 9.44 Undergraduate Econometrics, 2nd Edition –Chapter the Chow test is that it does not require the construction of the dummy and interaction variables Slide 9.45 Undergraduate Econometrics, 2nd Edition –Chapter Exercise 9.1 9.2 9.5 9.6 9.8 Slide 9.46 Undergraduate Econometrics, 2nd Edition –Chapter ... interaction variables These are variables formed by multiplying two or more explanatory Slide 9.3 Undergraduate Econometrics, 2nd Edition Chapter variables together When using either dummy variables. .. $1,649.17 Slide 9.18 Undergraduate Econometrics, 2nd Edition Chapter 9.5 Common Applications of Dummy Variables In this section we review some standard ways in which dummy variables are used Pay... interaction variables, some changes in model interpretation are required We will discuss each of these scenarios Slide 9.4 Undergraduate Econometrics, 2nd Edition Chapter 9.2 The Use of Intercept Dummy